Junliang Ye Shenghao Xie Ruowen Zhao Zhengyi Wang Hongyu Yan Wenqiang Zu Lei Ma Jun Zhu

Abstract

3D object editing is essential for interactive content creation in gaming,animation, and robotics, yet current approaches remain inefficient,inconsistent, and often fail to preserve unedited regions. Most methods rely onediting multi-view renderings followed by reconstruction, which introducesartifacts and limits practicality. To address these challenges, we proposeNano3D, a training-free framework for precise and coherent 3D object editingwithout masks. Nano3D integrates FlowEdit into TRELLIS to perform localizededits guided by front-view renderings, and further introduces region-awaremerging strategies, Voxel/Slat-Merge, which adaptively preserve structuralfidelity by ensuring consistency between edited and unedited areas. Experimentsdemonstrate that Nano3D achieves superior 3D consistency and visual qualitycompared with existing methods. Based on this framework, we construct the firstlarge-scale 3D editing datasets Nano3D-Edit-100k, which contains over 100,000high-quality 3D editing pairs. This work addresses long-standing challenges inboth algorithm design and data availability, significantly improving thegenerality and reliability of 3D editing, and laying the groundwork for thedevelopment of feed-forward 3D editing models. ProjectPage:https://jamesyjl.github.io/Nano3D